For the localization challenge, we provide three different scenarios, each consisting of a reference map and a query sequence from the 4Seasons dataset.
- City Loop
- reference map:
city_loop_2_train - query sequence:
city_loop_2_test
- Old Town
- reference map:
old_town_2_train - query sequence:
old_town_3_test
- Parking Garage
- reference map:
parking_garage_2_train - query sequence:
parking_garage_1_test
Each reference map contains highly accurate reference poses. All sequences from the 4Seasons dataset can be used as training data.
To download the challenge data and other sequences from the dataset, we refer to the download page.
For details on the dataset structure, we refer to the following documentation.
In this repository, you will find a set of tools for working with the 4Seasons dataset.
Each reloc_source_to_target.txt contains a list of first the keyframe from the source (reference) sequence and the query keyframe. The task is to provide the relative pose between the source and the query frame. Note that the relative pose is from cam0 of the reference sequence to cam0 of the query sequence, respectively. The 6DOF poses are specified as translation (t_x, t_y, t_z), and quaternion (q_x, q_y, q_z, q_w).
For the test sequences, the ground truth is withheld. For submitting your results to the challenge, please refer to the steps described at: https://sites.google.com/view/mlad-eccv2022/challenge.